jhadida / deck Goto Github PK

It might be better to move the folder of templates to the install location along with the Python scripts.
This would allow the user to define their own templates in a more sensible place than +dk/+mapred/templates.

Not sure about this. But if going through with it, we need to keep track of where the stuff is installed in order to initialise painlessly, so maybe a .bindir file created and ignored in the mapred folder.

Check if there is an issue with spaces in the paths provided to obj.Compiler. If so, find the correct way to escape them (ie either quote or escape spaces themselves).

Let mapfun behave as arrayfun on all non-cell inputs, including scalar structs.
Bring dk.struct.kvfun to first level in order to get a map function on struct or struct-arrays.

• Remove +verb/ submodule, in favour of using the dk.logger.Logger implementation.
• Default logger should be created at startup.
• Additional function dk.logger.level to simplify level control.
• Alias method Logger:setFile, more explicit than open.
• Implement static class dk.logger.test to be called by assert/reject in order to redirect assertions to different channels.
• Remove static method callerInfo from Logger, and move it to separate dk.logger.chan instead (taking stack depth argument), which returns a struct with channel ID and caller string.
• Update Logger:write to take a channel as first input, and exclude empty caller from log line.
• Implement function dk.log, taking a channel as first input (or creating a default one if formatted message is passed directly), with a persistent instance of the default logger.

funciton log(varargin)
    persistent..

    args = varargin;
    if nargin  > 1
        v = args{1};
        if isstruct(v)
            chan = v;
            args = varargin(2:end);
        elseif ismember(v,{'w','e','...'})
            chan = dk.logger.chan( v, 4 );
            args = varargin(2:end);
        else
            chan = dk.logger.chan( 'i', 4 );
        end
    end
    L.write( chan, args{:} );
end

• Refactor functions dk.warn/info to call dk.log
• Remove ..if methods from Logger, in favour of directed assertions.
• Replace calls to dk.wassert/wreject with calls to assert with specified channels.

Refactor struct submodule to support subfield assignment by default (i.e. using dk.str.to_substruct).

Is this code available under an open source license? I'm in the market for a new JSON parser...

I've found the following causes an error:

test = struct('value', {{1 2; 3 4}});
g = dk.json.encode(test);

Error using dk.json.encode>write_struct (line 46)
Assertion failed.
Error in dk.json.encode>write_array (line 100)
    s = write_struct(x,w,w.shapepfx);
Error in dk.json.encode>write_value (line 16)
            s=write_array(x,w,t);
Error in dk.json.encode>write_struct (line 57)
        s{i} = [w.ind w.enc.char([p f]) ':' w.sp write_value(x.(f),w)];
Error in dk.json.encode>write_scalar (line 128)
            s=write_struct(x,w);
Error in dk.json.encode>write_value (line 29)
            s=write_scalar(x,w,t);
Error in dk.json.encode (line 4)
    txt = write_value(dat,fmt); 

No issues with column or row cell arrays.

test = struct('value', {{1 2}});
g = dk.json.encode(test);
test = struct('value', {{1; 2}});
g = dk.json.encode(test);

Edit:
The issue is coming from

x = struct( 'size', dim, 'stride', str, 'value', reshape(x,1,numel(x)) );

In MATLAB struct('value', {1 2 3 4}) will create an array of structures (1x4 in this case). To get a structure containing a cell array you would need to double up the {}, so struct('value', {{1 2 3 4}}).
A possible solution could be:

    if strcmp(name, 'cell')
        x = struct( 'size', dim, 'stride', str, 'value', {reshape(x,1,numel(x))} );
    else
        x = struct( 'size', dim, 'stride', str, 'value', reshape(x,1,numel(x)) );
    end

Description

From Matlab version 2016a, the named argument npoints has changed to NumPoints without backwards compatibility.

To-do

Create a backwards-compatible function in dk.math and replace all occurences throughout the package.